Towards intelligent, adaptive input devices for users with physical disabilities

This thesis presents a novel application of user modelling, the domain of interest being the physical abilities of the user of a computer input device. Specifically, it describes a model which identifies aspects of keyboard use with which the user has difficulty. The model is based on data gathered in an empirical study of keyboard and mouse use by people with and without motor disabilities. In this study, many common input errors due to physical inaccuracies in using keyboards and mice were observed. For the majority of these errors, there exist keyboard or mouse configuration facilities intended to reduce or eliminate them. While such facilities are now integrated into the majority of modem operating systems, there is little published data describing their effect on keyboard or mouse usability. This thesis offers evidence that they can be extremely useful, even essential, but that further research and interface development are required. This thesis presents a user model which focuses on four of the most commonly observed keyboard difficulties. The model also makes recommendations for settings for three keyboard configuration facilities, each of which tackle one of these specific difficulties. As a user modelling task, this application presents a number of interesting challenges. Different users will have very different configuration requirements, and the requirements of individual users may also change over long or short periods of time. Some users will have cognitive impairments. Users may have very limited time and energy to devote to computer use. In response, this research has investigated the extent to which it is possible to model users without interrupting the task for which they are using a computer in the first place. This approach is appealing because it does not require users to spend time participating in model instantiation. This focus on inference rather than explicit testing or questioning also allows the model to dynamically track an individual user's changing requirements. This thesis shows that within the context of the keyboard difficulties studied, such an approach is feasible. The implemented model records users' keyboard input unintrusiveiy as they perform their own input tasks. This input is examined for evidence of certain types of input error or indications of difficulties in using the keyboard. In the model presented, conclusions are based on the assumption that the user is typing English text in a word processing application. However, the design of the model allows any other textual language to be used. A second empirical study, evaluating the model, is described. The model is shown to be very accurate in identifying users having difficulties in each of the areas tackled, the only exception being those who find a given operation awkward, but are able to perform it accurately. Where it is also possible to evaluate the configuration recommendations made by the model, the chosen settings are effective in reducing input errors and increasing user satisfaction with the keyboard. The model is also able to draw conclusions quickly for users with higher error rates, and shows good overall stability. In the light of this successful identification of keyboard difficulties, potential applications of the model are suggested. It could be used to help occupational therapists and assistive technologists to assess the keyboard configuration requirements of a new user. It could also be made available to users themselves - many people are currently unaware of facilities they may find useful, and how to activate them. The model could be extended to other areas of keyboard use, and to other input devices. This would allow systems to provide automatic, dynamic support for configuration, which would go some way towards improving the accessibility of computer systems for people with motor disabilities